8/11/2019 Talk Neural Coding ECVW Skye 04
1/30
Integrating neural coding into
cognitive models
Mike Oram
School of Psychology
University of St Andrews
8/11/2019 Talk Neural Coding ECVW Skye 04
2/30
Framework of research
8/11/2019 Talk Neural Coding ECVW Skye 04
3/30
8/11/2019 Talk Neural Coding ECVW Skye 04
4/30
Attention and neuronal codes
8/11/2019 Talk Neural Coding ECVW Skye 04
5/30
Fine temporal response measuresIdentifying repeating triplets Behavioural relevance
8/11/2019 Talk Neural Coding ECVW Skye 04
6/30
8/11/2019 Talk Neural Coding ECVW Skye 04
7/30
Neural codes in visual system
Coarse/medium (20-500ms resolution) Plenty of evidence for spike count
Evidence of medium (Richmond)
Fine temporal measures (1-2ms resolution) Synfire chains (includes synchrony)
We find no evidence (LGN, V1, Motor)
Even with behavioural relevance (TE: Oram, Lui, Richmond, in prep)
Synchrony
No evidence (TE/STS, Rolls et al. 2003, 2004)
8/11/2019 Talk Neural Coding ECVW Skye 04
8/30
8/11/2019 Talk Neural Coding ECVW Skye 04
9/30
8/11/2019 Talk Neural Coding ECVW Skye 04
10/30
Neural codes in late visual system
Evidence for spike count and response
latency Information encoded by spike count, latency when
Do these codes mean anything?
Spike counts Micro-stimulation studies (e.g. Newsome, Parker)
Latency
How to test? Cant stimulate to change latency
Want to relate to behaviour
8/11/2019 Talk Neural Coding ECVW Skye 04
11/30
Cognitive models of decisions
Information acquisition
hypothesis Loftus et al. Q J Exp
Psychol, 1983
Decisions (behaviouralRT) made wheninformation reachesthreshold
Carpenter & Williams,Nature 1995
Gold & Shadlen, TICS,2001
Carpenter & Williams,Nature, 1995
8/11/2019 Talk Neural Coding ECVW Skye 04
12/30
Modeling a decision
Use simulated neuralresponses to generate
population responses
Individual cellresponse profile fromOram & Perrett 1992
8/11/2019 Talk Neural Coding ECVW Skye 04
13/30
Modeling a decision
Cell tuning anddistribution of cellspreferred view(Perrett et al. 1991)
M d lli d i i
8/11/2019 Talk Neural Coding ECVW Skye 04
14/30
Modelling a decision
Use SCM to generate
responses of individualneurones
Use Fano factor of 1.5
8/11/2019 Talk Neural Coding ECVW Skye 04
15/30
Modeling a decision
Sum individual responses
to get population response
to target stimulus
8/11/2019 Talk Neural Coding ECVW Skye 04
16/30
8/11/2019 Talk Neural Coding ECVW Skye 04
17/30
Modeling a decision
8/11/2019 Talk Neural Coding ECVW Skye 04
18/30
Modeling a decision
Do as simplest computation Mutually inhibitory populations
Also done as integrate & fire neurones in
simple network (van der Meer, van Rossum, Edinburgh) Get same results as the poor mans version
Poor mans version MUCH quicker
What do neural codes predict for RT?
8/11/2019 Talk Neural Coding ECVW Skye 04
19/30
Both predict change in RT
Response Latency Response Magnitude
8/11/2019 Talk Neural Coding ECVW Skye 04
20/30
Both give
RT
changes
Unless can
delay thestart of
counting
Response Latency Response Magnitude
Count from 0 ms
Count from 150 ms
Modelling magnitude
8/11/2019 Talk Neural Coding ECVW Skye 04
21/30
Modelling magnitude
Modelling contrast
8/11/2019 Talk Neural Coding ECVW Skye 04
22/30
Modelling contrast
Predicting mean RT
8/11/2019 Talk Neural Coding ECVW Skye 04
23/30
Predicting mean RT
8/11/2019 Talk Neural Coding ECVW Skye 04
24/30
Testing mean RT predictions
Used a dual task experimental paradigm
Auditory task
odd or even number of pips
Visual task
mental rotation experiment with contrast
changes
8/11/2019 Talk Neural Coding ECVW Skye 04
25/30
Testing mean RT predictions
Is the letter normal or mirror form?
Used G and R
Testing mean RT predictions
8/11/2019 Talk Neural Coding ECVW Skye 04
26/30
Testing mean RT predictions
Oram et al.Phil Trans R Soc 2002
8/11/2019 Talk Neural Coding ECVW Skye 04
27/30
Testing RT distributions
Changing orientation -> skewing of distribution
8/11/2019 Talk Neural Coding ECVW Skye 04
28/30
Testing RT distributions
Changing contrast -> shifting of distribution
8/11/2019 Talk Neural Coding ECVW Skye 04
29/30
Testing RT distributions
Changing size -> skewing of distribution
Neural codes & cognition
8/11/2019 Talk Neural Coding ECVW Skye 04
30/30
Neural codes & cognition
Neural codes in TE/STS
Spike count and latency
Implications of neural codes for cognition
Predicts dual-task results Change response latency:
RT depends on SOA (Mean~SD relationship changes)
Change response magnitude: RT independent of SOA (Mean~SD relationship
constant)